[nsp] CAR

[Volodymyr Yakovenko]

Hello!

 I have had an argument with Cisco engineer (sorry, no names) concerning
 shaping issues recently. 

 The engineer states that applying CAR-limits to transit IP traffic is enough 
 good to effectively slow down TCP sessions speed.

 According to

http://www.cisco.com/en/US/tech/tk543/tk545/technologies_tech_note09186a00800a3a25.shtml

 and RFCs such point of view is questionable.

 Does anyone know any good information sources which investigate current TCP 
 stack's ability to recover after sudden packet loss (caused by CAR) without
 explicit congestion notification (ICMP source quench)?

 I would be most grateful for some numbers which show channel usage efficiency 
 with and without explicit congestion notification in different situations.

 Could someone familiar with CSS comment - does CSS support any new kind of
 shaping/limiting capabilites comparatively to traditional IOSes?  

-- 
Regards,
Volodymyr.